AR-24F
Application of precision agriculture technology to define and manage nematodes and diseases of soybean
John Rupe, Terry Kirkpatrick, Sreekala Bajwa, and Rick Cartwright, University of Arkansas and the Arkansas Cooperative Extension Service
Summary
Soybean Cyst Nematode (SCN) is a difficult-to-detect-threat to Arkansas soybean production. The objectives of the projects were to detect the onset and map the development of SCN in soybean fields using remote sensing and to determine the practicality and effectiveness of applying site- specific control measures for SCN. SCN counts at planting and harvest were collected for 2003 and in 2004 SCN at planting and at flowering stage were collected. Airborne color-infrared images were also acquired for both years. Geographically weighted regression (GWR) analysis was performed for the 2003 data. Ground level canopy multi-spectral data were recorded for 2004 and critical wavelengths, responsive for SCN counts were found by Maximum R2 procedures. In 2003, the yield with SCN counts, and soybean yield reflected a negative trend with nematode levels, showing higher values at lower cyst densities and lower values corresponding to high cyst densities for Hutcheson. Linear regression analysis showed significant spatial relationship between pre-planting SCN counts and yield for Hutcheson treatment (R2=0.59). A GWR model on SCN population densities at planting where nematodes appeared could explain 37% of variability in soybean yield. Analysis of spectral data with MaxR2 method resulted in four critical wavelengths (777.0, 779.5, 809.5 and 839.5 nm) in near infrared and one wavelength (269.5) in ultraviolet region with a model R2 value of 0.87. These wavebands were most responsive to SCN counts. The 2003 treatments had a significant impact on 2004 yield, with the resistant cultivar producing the highest yield followed by nematicide treatment and susceptible cultivar. The results promises use of remote sensing as an effective tool to detect the SCN in field, however, we recommend additional experiments, as we had only one data set of hyperspectral radiometric data.
Importance of Research
Soybean Cyst Nematode (SCN) is the most serious pest to the crop (Soybean Cyst Nematode Management Guide, 2004). Researchers (Zacharias et al., 1986; Bullock et al., 2000; Nutter et al., 2002) in the United States have used site-specific management strategies for detection and control of SCN. Damage due to SCN is difficult to detect as it causes aboveground symptoms similar to water stress. Plant leaves shrink and become yellow. Other than that, SCN effects are not often dramatic (Soybean Cyst Nematode Management Guide, 2004).
Although SCN-resistant cultivars have been developed, new races of SCN build up that are capable of damaging the resistant cultivars (Redcliffe et al., 1990). Rotation of crops has proved to be immensely beneficial; however, it is a long and non-profitable option for the soybean growers as they gain substantial profits only for alternate years (Giesler et al., 2002). Therefore growers are reluctant to use crop rotation. Few effective chemicals exist for controlling SCN (Levene et al., 1998). Applications of nematicide-herbicide-combination are found to be successful. However, insufficient diffusion of nematicide in clay soil prevents its wide use. The cost for blanket treatment of nematicide is unaffordable to most of the farmers (Levene, 1998).
The nematodes are unevenly distributed across the field, but they are aggregated. Site-specific application of inputs such as nematicide or planting of resistant cultivars seedling in soybean field is plausible if spatial and temporal distribution of SCN is clearly understood and structured. Unfortunately, success to attempts of investigating geostatistical distribution pattern has not been promising. However, Avendano et al., (2003) were able to identify locations of low and high cyst densities in the field between the years repeatedly by using geospatial tools such as semivarigrams, krigging and cross correlograms on log-transformed values of SCN data. They were not able to understand why nematodes appeared in clumps. In another approach, using the geospatial tools mentioned above, Avendano et al. (2004) found that soil texture is an important factor to explain the variability of SCN within infested fields. They found that SCN densities were higher in loamy sand than loamy clay soil.
Remote sensing can be used as a crop-monitoring tool and to understand within - field variability (Kollenkark et al., 1982). Landsat imagery was used to study reflectance of soybean canopies as affected by difference in row width, population, planting date and soil type. Kollenkark et al. (1982) found that changes in these agronomic variables manifested in canopy reflectance. In late 90’s researcher used remote sensing and GIS technology as an approach for site-specific SCN management (Nutter et al., 2001; Nutter et al., 2002; Noel et al., 1998) Nutter et al. (2002), found that strong statistical relationship (R2= 0.48) existed between SCN population density and ground-based percentage reflectance of the canopy sensed by spectroradiometer at 810 nm on a single date (July 13 in their study) for a SCN susceptible cultivar. Aerial image intensities values at 810 nm explained 33% variation in SCN population density and satellite image intensity explained 58 % variation in initial SCN population density. Carter (2001) contradicted Nutter et al. (2002), stating that in her study of small plots, initial SCN population showed very poor response to the percentage reflectance in 806 to 811 nm for both SCN resistant and susceptible cultivars.
Smolik et al. (2002) supported soil sampling of SCN infested areas guided by aircraft and satellite images (IKONOS) to detect qualitatively the damage caused by SCN and with this method they were able to note the progression of SCN symptoms over the growing season. No immediate alternative to labor intensive and monetarily expensive soil and plant sampling is available to detect the cause of stress in soybean. However, efforts were made to reduce extensive soil sampling with remote sensing (Rupe et al., 2003) in Arkansas.
Project Objectives
- To detect the onset and map the development of Soybean Cyst Nematode in soybean fields using aerial remote sensing
- To determine the practicality and effectiveness of applying site- specific control measures for SCN
Materials and Methods
Field experiments
Field experiments were conducted in 2004 at the Pine Tree experimental station, in Colt, Arkansas. The whole experimental field was planted with a high yielding SCN susceptible cultivar, Hutcheson. Soybean was planted on May 28 and harvested on November 11. These experiments are part of the project started in 2003. In 2003, a completely randomized block design were implemented in the field with 3 treatments which included replicated strips of a SCN susceptible cultivar, Hutcheson (Treatment A), Hutcheson + nematicide (Treatment B), and a SCN resistant cultivar, Anand (Treatment C). The strips were 16 rows wide and each treatment was replicated five times. Each row was 0.76 m wide and about 300 m in length. Each strip was divided into 8 equal grids resulting in a total of 120 quadrants. The field was planted on June 4, in 2003, and May 28, in 2004. The same grid layout was used in 2004 for field data collection and the only treatment was planting Hutcheson in the field.
SCN counting
Soil was sampled from each of the 120 grids. Nematodes were washed from soil and counted under microscope. Soil sampling was done three times, at planting, at mid-season and at harvest in 2004 for SCN counting. The sampling at harvest could not be completed as heavy rains kept the field wet until the end of this year and subsequent snowfall has delayed soil sampling.
Acquisition of reflectance at ground level
Soybean canopy reflectance was recorded in 2004 on July 17, August 14 and August 28 with a spectroradiometer (EPP 2000C, StellarNet Inc., Oldsmar, FL.) with a wavelength range of 200 to 882.5 nm and resolution of 0.5 nm. The data were recorded from a height of about 2 m from ground. Spectroradiometer was connected to a Toshiba (PROTÉGÉ 3020 CT) notebook computer via a Quatech PCMCIA card (Quatech Inc., Hudson, OH). The real time spectral graphing and recording was done with SpectraWiz software developed by StellarNet Inc., loaded on the notebook computer. Spectral data were collected on clear sunny days, between 1000 hrs to 1400 hrs, to minimize variability due to azimuth. Integration time of the spectroradiometer was set according to the light conditions, such that the bell shape curve of down welling radiance in scope mode was 90% of full screen. Light reflected by a white standard (Teflon) panel was recorded as reference irradiance. Canopy radiance was recorded at the center of each grid and three measurements were recorded for each grid. Canopy reflectance has calculated as the ratio of canopy radiance to Teflon panel radiance.
Vegetative Indices (VI) Used
Two vegetative indices namely, Normalized Difference Vegetation Index (NDVI and Green NDVI were calculated from reflectance values. The three values for vegetative indices corresponding to the three observations in a grid were arranged to obtain a mean VI for each plot. An average value of each index for three observations in a quadrant represented spectral index.
NDVI
The NDVI is the most commonly used vegetation index, although many vegetation indices have been developed to characterize green vegetation. The NDVI utilizes reflectance of the canopy in the near infrared (NIR) and red (R) bands of the spectrum and is determined by,
…………………… …………………………………..(1)
Green NDVI
The Green NDVI was calculated from the mean green reflectance between 565 and 575 nm (G) and mean NIR reflectance from 865 to 875 nm. Gitelson et al. (1996) found that this index was more sensitive to chlorophyll-a concentration. The Green NDVI is calculated as,
……………………………………………...... (2)
In the present study NDVI, GNVI) and the temporal changes in these indices over the season were tested for their relationship with SCN count and yield.
Acquisition of aerial photographs
Aerial color-infrared images were acquired on June 06, July 05, July 18, August 12, August 27 in 2004. In 2003, on July 11 and September 24 partial color infrared images were acquired. The first image acquired in June was a bare soil image. A four-band (550, 650, 750, 850 nm) camera (Model: MCA, Tetracam Inc., CA) type camera was mounted in the belly hole of Cessna 182 aircraft facing towards ground to acquire the images.
Soybean Yield
In 2003, yield data was recorded using a yield monitor on a mechanical harvester. Soybean yield was recorded for 2004 for each plot.
Maps and Statistical Analysis
ArcView GIS 3.2a and Desktop ArcGIS 8.2 software packages were used in preliminary analysis of spatial data layers generated for yield and SCN count at planting and harvest. Both the software packages are developed by Environmental Systems Research Institute (ESRI) Redlands, CA.
The 2003 data were analyzed using linear regression to investigate possible statistical links between SCN counts on planting and harvest and average yield for the quadrants where SCN were present. Relationship between vegetation indices and SCN counts was modeled with regression analysis using general linear model (GLM) procedures in SAS ( SAS Systems, Cary, NC) . The vegetation indices were calculated from spectral data collected on each date. The difference between each vegetation index for two consecutive dates was also calculated as a measure of temporal changes in canopy.
Geographically Weighted Regression Analysis
Geographically weighted regression analysis was performed by running GWR 3.0 software developed at the Department of Geography, University of Newcastle, UK (Fotheringham et al., 1996) to investigate statistical links between SCN count at planting and at harvest with the soybean yield averaged for the quadrant in which SCN were present. For a conventional or global linear regression applied to spatial data, we assume a stationary process as shown in equation 3.
y = b0 + b1x1 + b2x2 +… bnxn + e ……………………………….…….(3)
where b0, b1……bn are coefficients of regression, x1, x2, x3, …, xn are independent variables, y is the dependent or response variable, and e is the model error.
The parameter estimates obtained through the calibration of a global regression model are constant over space. It means that any spatial variations in the processes can be measured only by the error term. Spatial variability and patterns might be determined only by mapping the residuals, as the model does not account for the spatial variability. In other words, simple linear regression does not account for spatial correlation in the data.
The GWR methodology accounts for the spatial collinearity in data with the use of distance-weighted sub-samples of the data to produce local linear regression estimates for every point in space (Brunsdon et al., 1996). Each set of parameter estimates is based on a distance-weighted sub-sample of neighboring observations. The general format id a GWR model is
y(g) = b0(g) + b1 (g) x1 + b2 (g) x2 +… bn (g) xn + e (g)………(4)
where (g) refers to a location at which estimates of the parameters is performed. Although this model accounts for the spatial variability in the best manner, it has some drawbacks associated with it. One problem with GWR estimates is that valid inferences cannot be drawn for the regression parameters by traditional least squares approaches. The other problem is that the locally linear estimates based on a distance-weighted sub-sample of observations may suffer from weak data at some locations. We have made an attempt to establish relationship between the SCN counts at planting and harvest, and spectral indices (NDVI, and Green NDVI) derived from single date airborne image soybean canopy covered under each treatment. We assumed that the true impact of SCN infestation on spectral reflectance can be captured only if we account for the spatial collinearity between treatments.
Maximum R2 procedures
For 2004 spectroradiometer data wavelengths, which were most responsive to SCN were found by Maximum R2 procedure in SAS (SAS Systems, Cary, NC) among all the wavelengths in the range of 247 nm to 882.5 nm for data in 2004. Wavelength was an independent variable and SCN count was treated as the dependent variables.
In case of maximum R2, the model selection procedure starts by selection of a single variable model and generating R2 for it. The model then starts accommodating variables one by one until no other varible enhances R2 (SAS Systems, Cary, NC). The resulting model could then be a two variable or even a six-variable model. In the analysis presented, the iterations were limited to 5.